About Scovant

We're building the measurement system for the agentic web.

For thirty years, the web's been built for humans with eyeballs and impatience. The next ten years it gets used by something quieter: agents acting on behalf of those humans. Same browsers, completely different failure modes — and almost nobody is measuring.

The entity card

What is Scovant?

Scovant is an AI agent readiness testing platform that scans, simulates, and continuously monitors websites to determine how reliably AI agents can discover, understand, and interact with them.

How Scovant works

01
Scan
A full-site crawl evaluates the documented compatibility rule set and produces a public, reproducible score.
02
Simulate
AI agents attempt real tasks — add to cart, book, check out — across multiple LLM models.
03
Monitor
CI hooks and scheduled re-scans compare every change against a baseline and flag agent regressions.

Terminology

What Scovant evaluates

AI crawler accessibility
Whether AI crawlers and agents can reach the site at all — robots.txt policy per bot, WAF/CDN behavior, and whether the declared policy matches what actually happens.
Structured data
Whether the content and actions of a site are exposed in machine-readable form agents can rely on, not just rendered pixels.
Agent protocol discovery
Detection of agent-facing interfaces the site declares — MCP, WebMCP, and UCP — and whether they behave as declared.
Browser-agent task simulation
Real AI agents drive a real browser through actual purchase and booking flows, so the result is what happens, not what is declared.
CI regression detection
Every deploy is compared against a baseline, so a change that breaks an agent task is caught before agents hit it in production.

Official resources

Website
scovant.com
Research
The Agent Web Index — open monthly dataset
Documentation
Guides and reference
Scoring methodology
Weights, statuses and thresholds, published
Rule catalog
Every compatibility rule we evaluate
Glossary
Canonical term definitions
Comparisons
How Scovant relates to adjacent categories
Scovant Core (open source)
Scovant Core measures passive, machine-facing signals. Scovant Cloud verifies how real agents actually behave, across providers, browser runtimes, security layers and time.
Pricing
Plans and credit packs
The thesis

The agentic web needs a scoreboard.

You can't fix what you can't see, and you can't compare what you don't measure the same way every time. We think there should be one number — comparable, public, reproducible — for whether a site is going to work when an agent shows up.

01
Agents will route more commerce than you think.
Conservative analysts have agents at 8% of e-commerce by 2027. Less conservative ones, 30%. Either way, today the floor is broken.
02
The compatibility surface is small but specific.
There are about twenty things sites do that break agents, and ninety percent of teams haven't heard of half of them. Lighthouse doesn't check for them.
03
A score in the footer is going to matter.
Verifiable Agent Readiness ratings will become a trust signal. Like SOC 2, or HTTPS in the address bar — boring once everyone has one, embarrassing if you don't.
The landscape

How we got here.

2022
ChatGPT ships.
Suddenly a billion people are talking to a language model. Within a year, plugins, then function calling, then the words "agent" and "tool use" enter the chat.
2024
The first wave of browsing agents.
OpenAI's Operator, Anthropic's computer use, Adept, Browser-use, Manus. They can drive a browser. They mostly fail at checkout.
Q1 2026
Agent revenue is real but unmeasurable.
Agent-driven purchases are becoming real revenue — and almost no one can measure whether their own site actually works for agents.
May 2026
We start Scovant.
Five years from the agentic web being normal, zero years from anyone agreeing on what "agent-ready" means.
What we believe

How we work, in four sentences.

Boring is a feature.
A measurement tool has one job — be the same on Tuesday as it was last month. We pick deterministic rules over clever ones, every single time.
Transparent Score. Proprietary Engine.
Every score is explainable: the categories, their weights, and every rule are published, like a benchmark suite. The calibration layer — anti-gaming heuristics and exact cap mechanics — stays proprietary, and we say exactly which parts. We welcome corrections.
Verifiable reports.
Your reports are private by default. Publish one to the showcase (after verifying your domain) and it becomes a link-able, embed-able public report; anonymous free checks are never indexed.
Devs read the docs.
Our buyers are engineers and the team they work with. We don't hide pricing, gate features behind sales calls, or use “starting at” copy. Tools should explain themselves.

We'd love your site on our scoreboard.

Scan once. Tell us what we missed. Help us shape the rule set everyone ends up using.